The author of this idea used it to "accurately" predict the date of the fall of the Berlin wall. The basic idea is that if you know the age of something, there is a 50% chance that it will end after between one third and three times its current age. I checked the math; it is numerically correct.

However, when the prediction is right, the margins can be huge. If you estimated the lifetime of the Berlin Wall in, say, 1982, the model predicts it will fall sometime between 1989 and 2045. That's not useful.

The model also conflates the idea of being right 50% of the time, and having a 50% chance of being right at *any given* time. And, when its predictions are wrong, they're hilariously wrong.

Mathematically, this model works just as well for predicting the remaining height of a mountain, based on how far up it you have climbed. For a thousand foot mountain, the prediction will be correct 50% of the time (between 250 and 750 feet). However, at 999 ft, with *one foot* left to go, it will predict you have between 333 and 2997 feet left. Even assuming you're ascending the mountain in complete fog, and can't just look at the peak, or do trigonometry or something, the only assumption you can make after climbing the mountain X feet is that the mountain must be at least X feet high. The model isn't giving you anything of value.

More examples:

A 6-day old baby has a 50% chance of having a total lifespan between 8 and 24 days.

A car with 300,000 miles on it has a 50% chance of lasting another 100,000 to 900,000 miles.

Being right 50% of the time over the range of possible samples obviously does not mean that your model has a 50% chance of being right for any given sample.